Monkey patch uvicorn multi worker is_alive timeout (#10159)

Co-authored-by: Huang Long <121648372+llll114@users.noreply.github.com>

python/sglang/srt/entrypoints/http_server.py

@@ -93,6 +93,7 @@ from sglang.srt.managers.io_struct import (
 from sglang.srt.managers.multi_tokenizer_mixin import (
     MultiTokenizerManager,
     get_main_process_id,
+    monkey_patch_uvicorn_multiprocessing,
     read_from_shared_memory,
     write_data_for_multi_tokenizer,
 )
@@ -1219,6 +1220,9 @@ def launch_server(
                 "level": "INFO",
                 "propagate": False,
             }
+
+            monkey_patch_uvicorn_multiprocessing()
+
             uvicorn.run(
                 "sglang.srt.entrypoints.http_server:app",
                 host=server_args.host,

python/sglang/srt/managers/multi_tokenizer_mixin.py

@@ -19,6 +19,7 @@ import os
 import pickle
 import sys
 import threading
+from functools import partialmethod
 from multiprocessing import shared_memory
 from typing import Any, Dict
 
@@ -556,3 +557,17 @@ def write_data_for_multi_tokenizer(
     args_shm.close()
 
     return args_shm
+
+
+def monkey_patch_uvicorn_multiprocessing(timeout: float = 10):
+    """Monkey patch uvicorn multiprocessing is_alive timeout"""
+    # from default 5s -> 10s
+    try:
+        from uvicorn.supervisors.multiprocess import Process
+
+        Process.is_alive = partialmethod(Process.is_alive, timeout=timeout)
+
+    except ImportError:
+        logger.warning(
+            "uvicorn.supervisors.multiprocess not found, skipping monkey patch"
+        )